Allow ports to be reused in gloo #353
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
This pull request was exported from Phabricator. Differential Revision: D44029927 |
H-Huang
added a commit
to H-Huang/gloo
that referenced
this pull request
Mar 27, 2023
Summary: X-link: pytorch/pytorch#97677 Pull Request resolved: facebookincubator#353 ProcessGroupGloo and gloo seem to be opening and closing sockets without allowing the port to be reused. We see this issue pop up in larger training jobs "Address already in use" and we assume it to be because all the ephemeral ports are exhausted. This diff allows ports to be reused, we see a reduced number of ports being in `TIME_WAIT` state. context: https://fb.workplace.com/groups/319878845696681/permalink/5988899781205532/ another issue: https://fb.workplace.com/groups/319878845696681/permalink/958768178474408/ Differential Revision: D44029927 fbshipit-source-id: 447186248c900f122c56ae824e0a8eb56aab732f
|
This pull request was exported from Phabricator. Differential Revision: D44029927 |
H-Huang
force-pushed
the
export-D44029927
branch
from
191a459 to
1b3a5de
Compare
H-Huang
added a commit
to H-Huang/gloo
that referenced
this pull request
Mar 27, 2023
Summary: X-link: pytorch/pytorch#97677 Pull Request resolved: facebookincubator#353 ProcessGroupGloo and gloo seem to be opening and closing sockets without allowing the port to be reused. We see this issue pop up in larger training jobs "Address already in use" and we assume it to be because all the ephemeral ports are exhausted. This diff allows ports to be reused, we see a reduced number of ports being in `TIME_WAIT` state. context: https://fb.workplace.com/groups/319878845696681/permalink/5988899781205532/ another issue: https://fb.workplace.com/groups/319878845696681/permalink/958768178474408/ Differential Revision: D44029927 fbshipit-source-id: 31f1c91d40686001195bb8b03baec38b88dfbb12
H-Huang
force-pushed
the
export-D44029927
branch
from
1b3a5de to
942e7de
Compare
|
This pull request was exported from Phabricator. Differential Revision: D44029927 |
|
This pull request was exported from Phabricator. Differential Revision: D44029927 |
H-Huang
added a commit
to H-Huang/gloo
that referenced
this pull request
Mar 29, 2023
Summary: X-link: pytorch/pytorch#97677 Pull Request resolved: facebookincubator#353 ProcessGroupGloo and gloo seem to be opening and closing sockets without allowing the port to be reused. We see this issue pop up in larger training jobs "Address already in use" and we assume it to be because all the ephemeral ports are exhausted. This diff allows ports to be reused, we see a reduced number of ports being in `TIME_WAIT` state. context: https://fb.workplace.com/groups/319878845696681/permalink/5988899781205532/ another issue: https://fb.workplace.com/groups/319878845696681/permalink/958768178474408/ Differential Revision: D44029927 fbshipit-source-id: a12967a5c7120677a34efccb7da9bfc340f41a5f
H-Huang
force-pushed
the
export-D44029927
branch
from
942e7de to
4464366
Compare
H-Huang
added a commit
to H-Huang/gloo
that referenced
this pull request
Mar 29, 2023
Summary: X-link: pytorch/pytorch#97677 Pull Request resolved: facebookincubator#353 ProcessGroupGloo and gloo seem to be opening and closing sockets without allowing the port to be reused. We see this issue pop up in larger training jobs "Address already in use" and we assume it to be because all the ephemeral ports are exhausted. This diff allows ports to be reused, we see a reduced number of ports being in `TIME_WAIT` state. context: https://fb.workplace.com/groups/319878845696681/permalink/5988899781205532/ another issue: https://fb.workplace.com/groups/319878845696681/permalink/958768178474408/ Differential Revision: D44029927 fbshipit-source-id: 1f83e9288776a6ec6e2f2b1ea356739ae057d4a6
H-Huang
force-pushed
the
export-D44029927
branch
from
4464366 to
b832c03
Compare
|
This pull request was exported from Phabricator. Differential Revision: D44029927 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D44029927 |
H-Huang
added a commit
to H-Huang/gloo
that referenced
this pull request
Mar 29, 2023
Summary: X-link: pytorch/pytorch#97677 Pull Request resolved: facebookincubator#353 ProcessGroupGloo and gloo seem to be opening and closing sockets without allowing the port to be reused. We see this issue pop up in larger training jobs "Address already in use" and we assume it to be because all the ephemeral ports are exhausted. This diff allows ports to be reused, we see a reduced number of ports being in `TIME_WAIT` state. context: https://fb.workplace.com/groups/319878845696681/permalink/5988899781205532/ another issue: https://fb.workplace.com/groups/319878845696681/permalink/958768178474408/ Differential Revision: D44029927 fbshipit-source-id: c29af50177d52f17a98db73d7a63fc436708626f
H-Huang
force-pushed
the
export-D44029927
branch
from
b832c03 to
751f4fe
Compare
|
This pull request was exported from Phabricator. Differential Revision: D44029927 |
H-Huang
added a commit
to H-Huang/gloo
that referenced
this pull request
Mar 29, 2023
Summary: X-link: pytorch/pytorch#97677 Pull Request resolved: facebookincubator#353 ProcessGroupGloo and gloo seem to be opening and closing sockets without allowing the port to be reused. We see this issue pop up in larger training jobs "Address already in use" and we assume it to be because all the ephemeral ports are exhausted. This diff allows ports to be reused, we see a reduced number of ports being in `TIME_WAIT` state. context: https://fb.workplace.com/groups/319878845696681/permalink/5988899781205532/ another issue: https://fb.workplace.com/groups/319878845696681/permalink/958768178474408/ Differential Revision: D44029927 fbshipit-source-id: cf12f681fd1a42249ffac259d9d34ab076a5ac94
H-Huang
force-pushed
the
export-D44029927
branch
from
751f4fe to
463c263
Compare
|
This pull request was exported from Phabricator. Differential Revision: D44029927 |
H-Huang
added a commit
to H-Huang/gloo
that referenced
this pull request
Mar 30, 2023
Summary: X-link: pytorch/pytorch#97677 Pull Request resolved: facebookincubator#353 ProcessGroupGloo and gloo seem to be opening and closing sockets without allowing the port to be reused. We see this issue pop up in larger training jobs "Address already in use" and we assume it to be because all the ephemeral ports are exhausted. This diff allows ports to be reused, we see a reduced number of ports being in `TIME_WAIT` state. context: https://fb.workplace.com/groups/319878845696681/permalink/5988899781205532/ another issue: https://fb.workplace.com/groups/319878845696681/permalink/958768178474408/ Differential Revision: D44029927 fbshipit-source-id: 4a1483d0eceda01ffd02c7747282129f7f4a2efe
H-Huang
force-pushed
the
export-D44029927
branch
from
463c263 to
61f8d06
Compare
H-Huang
added a commit
to H-Huang/pytorch
that referenced
this pull request
Mar 30, 2023
Summary: Pull Request resolved: pytorch#97677 X-link: facebookincubator/gloo#353 ProcessGroupGloo and gloo seem to be opening and closing sockets without allowing the port to be reused. We see this issue pop up in larger training jobs "Address already in use" and we assume it to be because all the ephemeral ports are exhausted. This diff allows ports to be reused, we see a reduced number of ports being in `TIME_WAIT` state. context: https://fb.workplace.com/groups/319878845696681/permalink/5988899781205532/ another issue: https://fb.workplace.com/groups/319878845696681/permalink/958768178474408/ Test Plan: Add a gloo test to create 4 groups of size 64 using multithreaded PG + gloo. In total 256 ranks. Differential Revision: D44029927 fbshipit-source-id: 9c31c38485333602c33e12c12813bea33ccb9438
Summary: X-link: pytorch/pytorch#97677 Pull Request resolved: facebookincubator#353 ProcessGroupGloo and gloo seem to be opening and closing sockets without allowing the port to be reused. We see this issue pop up in larger training jobs "Address already in use" and we assume it to be because all the ephemeral ports are exhausted. This diff allows ports to be reused, we see a reduced number of ports being in `TIME_WAIT` state. context: https://fb.workplace.com/groups/319878845696681/permalink/5988899781205532/ another issue: https://fb.workplace.com/groups/319878845696681/permalink/958768178474408/ Differential Revision: D44029927 fbshipit-source-id: b531d67456d4656ce23e9db7c4cb892d8fc90475
H-Huang
force-pushed
the
export-D44029927
branch
from
61f8d06 to
4adb00a
Compare
|
This pull request was exported from Phabricator. Differential Revision: D44029927 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
ProcessGroupGloo and gloo seem to be opening and closing sockets without allowing the port to be reused. We see this issue pop up in larger training jobs "Address already in use" and we assume it to be because all the ephemeral ports are exhausted.
This diff allows ports to be reused, we see a reduced number of ports being in
TIME_WAITstate.context: https://fb.workplace.com/groups/319878845696681/permalink/5988899781205532/
another issue: https://fb.workplace.com/groups/319878845696681/permalink/958768178474408/
Differential Revision: D44029927